布尔型贝叶斯网络参数学习

Parameter Learning of Boolean Bayesian Networks

布尔型贝叶斯网络是一类由布尔型变量组成的网络,它能够以线性多变量函数描述,使计算和处理上灵活高效。通过运用连接树算法对络进行分块化处理的方法,可以提高算法的效率,然后以传统的最大似然估计方法对布尔型网络的参数进行学习。服从同一分布律的贝叶斯网络参数学习算法发展比较成熟,这类以狄利克雷或者高斯分布为基础的算法在应用领域中难以发挥其应有的价值。相比之下,基于布尔型贝叶斯网络下的参数学习更贴近于应用,在人工智能和数据挖掘等领域有很好的发展前景。

Boolean Bayesian network is a class of Bayesian networks which are made up of Boolean varia-bles. The method to describe the network with a multi-linear function is flexible and efficient to compute and process variables. By introducing Junction Tree algorithm,the network can be divided into blocks which can make it easy to calculate. Then the traditional maximum likelihood estimation method was used for learning Boolean networks. Parameter learning algorithm following the same distribution is more ma-ture,but this kind of algorithm based on Dirichlet or Gaussian distribution is difficult to play its proper val-ue in practice. In contrast,parameter learning based on Boolean networks gets close to applications. It has good prospects for development in areas such as artificial intelligence and data mining.